Imbibition Triggered by Capillary Condensation in Nanopores

Jan 25, 2017 - ... authors thank Glenn Swan for technical support and Antoine Robin for help in building the vacuum system. This work was supported by...
0 downloads 6 Views 1MB Size
Download PDF
Article pubs.acs.org/Langmuir

Imbibition Triggered by Capillary Condensation in Nanopores Olivier Vincent,*,† Bastien Marguet,† and Abraham D. Stroock*,†,‡ †

Robert Frederick Smith School of Chemical and Biomolecular Engineering and ‡Kavli Institute at Cornell for Nanoscale Science, Cornell University, Ithaca, New York 14853, United States ABSTRACT: We study the spatiotemporal dynamics of water uptake by capillary condensation from unsaturated vapor in mesoporous silicon layers (pore radius rp ≃ 2 nm), taking advantage of the local changes in optical reflectance as a function of water saturation. Our experiments elucidate two qualitatively different regimes as a function of the imposed external vapor pressure: at low vapor pressures, equilibration occurs via a diffusion-like process; at high vapor pressures, an imbibition-like wetting front results in fast equilibration toward a fully saturated sample. We show that the imbibition dynamics can be described by a modified Lucas−Washburn equation that takes into account the liquid stresses implied by Kelvin equation.

■

INTRODUCTION The interaction of porous media with liquids and vapors occurs in many contexts. In nature, water is ubiquitous within soils, rocks, or the atmosphere, and its behavior in pores is fundamental for the hydration of plants and their vascular flows,1 the stability of sand terrains,2 and the formation of clouds by condensation on atmospheric particles.3 In technology, the permeation of liquids through porous media occurs in a diversity of applications, from printing ink on paper to oil recovery.4 As a result, the dynamics of invasion of a pore space when put in contact with liquid, i.e. imbibition, has been the subject of many studies, from the early works of Lucas and Washburn5,6 to recent works on nanoporous media.7,8 The observed dynamics of invasion show excellent agreement with the classical Lucas−Washburn equations, even in pores approaching the molecular size.9,10 Nanopores, however, can fill with liquid even if in contact with the vapor phase only. This phenomenon, known as capillary condensation, occurs for any pore size but persists for vapor pressures significantly below saturation only for diameters in the nanometer range.11 Capillary condensation plays a role in many areas of science and technology such as the cohesion and friction of granular materials,12 or the dynamics of hydrocarbons in subsurface reservoirs.13 Adsorption by capillary condensation, and the reverse process of desorption are still subject of active research, with outstanding questions related to the collective effects induced by disorder in the pore network14,15 or to deviations from macroscopic thermodynamics and dynamics at the nanoscale.10,16 The experimental study of both imbibition and capillary condensation at the nanoscale is challenging due to the difficulty of accessing the local filling state of the porous medium: often only the global mass uptake can be recorded.17,18 Spatially resolved information can however be obtained by more sophisticated techniques such as interferometry,7 neutron radiography8 or capacitance measurements.19 Here, we use © XXXX American Chemical Society

the local changes in reflectance of quasi two-dimensional, laterally connected porous silicon layers to study the dynamics of water uptake as a function of the imposed water vapor pressure (relative humidity) around the medium. We elucidate a qualitative transition as a function of humidity from a diffusion-like regime to an imbibition-like regime. For the imbibition-like regime, we show that a modified Lucas−Washburn equation that takes into account the effective capillary pressure predicted from Kelvin− Laplace equation explains our observations quantitatively.

■

METHODS

We formed a 5 μm thick layer of mesoporous silicon (PoSi) on top of a p-type silicon (Si) wafer of ⟨111⟩ crystal orientation and 1−10 Ω·cm resistivity, by anodization in a mixture of 1:1 49% hydrofluoric acid/ pure ethanol at 20 mA/cm2 current density for 5 min. We then used thermal oxidation at 700 °C in pure oxygen for 30 s to increase hydrophilicity and stabilize the porous layer. We expect a laterally connected pore structure with porosity, ϕ = 0.45 and a typical pore radius, rp = 1.5−2 nm, as estimated in previous studies from nitrogen sorption porosimetry and dynamic flow methods.10,20 The sample resulting from these processing steps is sketched in Figure 1a. Such a sample could equilibrate quickly with the environment due to the open top surface and to the small thickness of the porous silicon layer (∼μm). For that reason, we measured the reflectance isotherms (see below) at this step in the fabrication process. We then sealed the top surface to glass by anodic bonding (Figure 1b), so that the sample could exchange vapor only at its open edge; this change resulted in much slower equilibration dynamics in the y direction due to the large lateral dimensions of the porous layer (∼cm). Figure 1c presents a top-view photograph of a sample as in Figure 1b. We placed the sample in a vacuum chamber equipped with an optical window, thermostated at T = 15 °C (Figure 1d). Pure water vapor, obtained by evaporation from a degassed liquid water source, Received: December 19, 2016 Revised: January 22, 2017 Published: January 25, 2017 A

DOI: 10.1021/acs.langmuir.6b04534 Langmuir XXXX, XXX, XXX−XXX

Article

Langmuir

Figure 1. Overview of experiments. (a-b) Side-view sketch of the sample before (a) and after (b) bonding to glass. (c) Top-view photograph of the sample. The dashed square shows the portion of the sample displayed in Figure 3a. (d) Sketch of the vacuum system used to control the vapor pressure and the temperature. (e) Vapor pressure signal during an experimental run, showing a typical early transient. (f) Schematic of the process of capillary condensation of vapor into liquid in a pore. (g) Pore-scale representation of the dynamics of liquid invasion following condensation at high vapor pressure, driven by the competition between the capillary pressures P1 (determined by Kelvin equilibrium: eq 2) and P2 (determined by the geometry of the pore and the equilibrium contact angle of the liquid on the pore wall: eq 4). flowed through the chamber toward the vacuum pump, and we measured its pressure p using a pressure gauge connected to the chamber. By adjusting the relative opening of two needle valves v1 and v2, we could set the pressure between p = 0 and p = psat. All capillary condensation experiments started with a dry sample equilibrated for at least 24 h at p = 0 (v1 closed). Opening v1 at a time defined as t = 0 allowed us to impose a steady, nonzero vapor pressure p after a transient time of typically ≃10 s (Figure 1e). From the error of the pressure gauge reading as well as the measured fluctuations of T and p during the experiments, we estimated an uncertainty on the activity, a = p/psat of ±0.005. We recorded time-lapse image sequences of the sample, and analyzed its response to changes in vapor pressure by extracting the local relative change in reflectance ΔI/I under constant, white-light, diffuse illumination. From the images, we calculated (ΔI/I)i = ref (gref i − gi)/gi where gi was the grayscale value of the ith pixel in the was its average value for t < 0, image (gi ∈ [0−255]) and gref i corresponding to the dry sample. Water uptake produced a darkening of the image, resulting in (ΔI/I)i > 0. For the measurement of the reflectance isotherm, we evaluated the average change in intensity, 1 ⟨ΔI /I ⟩ = N ∑i (ΔI /I )i over the whole region of interest shown in Figure 1c (⟨ΔI/I⟩xy), while for the extraction of invasion dynamics propagating along y, averaging was done along the x direction only (⟨ΔI/I⟩x(y), axes shown in Figure 1c).

■

(P − psat) and μvap = μ0 + RT ln(a), where vm is the molar volume of the liquid and R is the universal gas constant. The expression of μvap uses the assumption that the vapor behaves as an ideal gas, while that of μliq is obtained assuming an incompressible liquid (vm independent of pressure, which results in negligible errors for water at room temperature10). Eventually, the equilibrium condition through the equality of chemical potential requires

P = psat + Ψ(a) where we have defined the vapor water potential Ψ(a) =

THEORY

p psat

RT ln(a) vm

(3)

Equation 2 is known as the Kelvin equation; it is based on macroscopic thermodynamic considerations and bulk descriptions of the pore fluid, but has been shown to accurately describe liquid capillary stresses due to subsaturated liquid−vapor equilibrium, even in nanoscale confinements.10,21,22 Liquid Invasion Dynamics. At high vapor pressure, it becomes thermodynamically favorable for the pores to fill with liquid by the process of capillary condensation.11 Capillary condensation causes the formation of liquid plugs at the sample edge (Figure 1f), which progressively invade the whole depth of the pores. We assume that the pore filling after capillary condensation proceeds by bulk liquid flow driven by the competition of capillary pressures P1 and P2 as depicted in Figure 1g, and that condensation of vapor at the edge (meniscus 1) provides the liquid necessary to account for the advance of meniscus 2. In other words, we consider that the pore wetting dynamics is slow compared to the dynamics of vapor condensation at the edge,23 so that meniscus 1 is in local thermodynamic equilibrium with the vapor. As a result, from eqs 2 and 3, P1 = Ψ(a), where we have neglected psat compared to Ψ. We assume that the curvature of meniscus 2 is defined by the local mechanical equilibrium of the triple line, i.e., the curvature is * = − 2cos θ /rp where θ is the equilibrium contact angle of the liquid on the solid and rp the radius of the pore. From Laplace equation, P2 − p = ΔPc where

Kelvin Equation. We consider liquid−vapor equilibrium at temperature T with liquid at pressure P and vapor at pressure p. It is convenient to define the activity, a, of the vapor, which compares the vapor pressure to its saturation value psat(T), a=

(2)

(1)

Note that the quantity a × 100 is also the percent relative humidity of the vapor [%RH]. Bulk equilibrium between liquid and vapor imposes P = p = psat(T) . In a porous medium, however, the liquid and vapor phase can coexist at different pressures due to the capillary pressure associated with the curved liquid−vapor meniscus. From the Laplace equation P − p = σ *, where * [m−1] is the curvature of the meniscus and σ [N/m] is the surface tension of the liquid. The condition for equilibrium, even with P ≠ p, is the equality of the chemical potentials μliq and μvap in the liquid and vapor phases. Integration of the isothermal Gibbs−Duhem equation, taking P = p = psat (chemical potential μ0) as a reference state, yields μliq = μ0 + vm

ΔPc = −

2σ cos θ rp

(4)

is an intrinsic pore capillary pressure. Neglecting the vapor pressure p compared to P2, we have P2 = ΔPc and

ΔP = P1 − P2 = Ψ(a) − ΔPc B

(5) DOI: 10.1021/acs.langmuir.6b04534 Langmuir XXXX, XXX, XXX−XXX

Article

Langmuir

Dm ≃ 8 × 10−7 m2/s for water vapor in pores 2 nm in radius, corresponding to a transport coefficient k ≃ 2 × 10−12 m2/s. Assuming α′ ∼ 1 for typical adsorption isotherms, we expect D to also be on the order of 10−12 m2/s if transport is based on Knudsen diffusion. This value will serve as a reference in our discussion of the diffusion mechanisms below.

The driving force for the flow is the pressure gradient ∇P. Considering that the invasion dynamics is slow compared to poroelastic transients,10 we assume that ∇P is uniform in space; the pressure gradient is thus simply ∇P = ΔP/Y(t), where Y(t) is the front position (Figure 1g). We neglect intertial effects and describe the flow response using Darcy equation q = ρliqκ∇P, where q [kg/(m2·s)] is the mass flux, ρliq is the density of the liquid, and κ [m2/(Pa·s)] is the permeability of the porous layer. Using the conservation of mass q = ρliqϕ dY/dt, the front position Y(t) obeys the differential equation Y dY/dt = κΔP/ϕ. Integration using Y = 0 at t = 0 yields

Y 2(t ) 2κ =w= (Ψ(a) − ΔPc) t ϕ

■

RESULTS Reflectance Isotherm. We first recorded the static changes in reflectance of the porous silicon layer as a function of the vapor activity a = p/psat with an open sample (not bonded to glass) as sketched in Figure 1a. The resulting ref lectance isotherm is presented in Figure 2. These reflectance isotherms

(6)

where we have used eq 5 to express the pressure difference ΔP. When Ψ(a) = 0 (i.e., p = psat from eqs 1 and 3), eq 6 reduces to the Lucas− Washburn law for the dynamics of imbibition, which describes the noninertial dynamics of liquid invasion when the sample edge is plunged in bulk liquid.24 The extra term Ψ(a) accounts for the existence of an additional capillary pressure governed by Kelvin equation when the sample is in contact with vapor instead of liquid (Figure 1g). This effect of Kelvin equation on liquid stresses and flow dynamics has been considered previously in other contexts, such as pervaporation and drying fluxes through nanoporous materials.10,25 We will refer to the parameter w as the imbibition speed coeff icient in the rest of the paper. We note here that we have not considered the additional stresses associated with the liquid−solid interaction in the pores. Such stresses can play an important role for the global deformations induced by sorption in nanoporous materials;26 however, it is unlikely that they impact liquid flow dynamics, because they should be both normal to the flow direction and uniform in space in the liquid-filled zone. As a result, we expect that their effect cancels out in the formulas governing the flow dynamics (e.g., eq 6). Vapor Invasion Dynamics. At low vapor pressures, no capillary condensation occurs, but vapor can adsorb on the pore walls. The average density ρ(a) inside of the pore depends on the vapor activity, and varies between ρ = 0 at a = 0 (evacuated sample) to ρ = ρliq, as a approaches 1 (pore filled with liquid of density ρliq due to capillary condensation). It is thus convenient to define α(a) = ρ(a)/ ρliq, which takes values between 0 and 1. The dynamics of vapor invasion in the pores is a result of the combined effects of transport through the pores and adsorption on the pore walls, and obeys the conservation of mass ∂q/∂y = −ϕ∂ρ/∂t. We define a generic transport coefficient k(a) so that q = −ϕρliqk(a)∂a/∂y; the conservation of mass then translates into

α′

∂a ∂⎛ ∂a ⎞ = ⎜k(a) ⎟ ∂t ∂y ⎝ ∂y ⎠

Figure 2. Reflectance isotherm of the sample measured on the open sample (Figure 1a). ⟨ΔI/I⟩xy represents the relative change in average grayscale intensity with respect to the dry sample (see Methods). ⟨ΔI/I⟩xy > 0 corresponds to a darker image and to a higher sample saturation.

display features characteristic of adsorption and desorption isotherms measured by mass in nanoporous media:11 a reversible plateau at high relative humidities (here for a > 0.65), associated with the presence of bulk liquid in all the pores; a reversible branch at low relative humidities (here for a < 0.4), associated with the presence of adsorbed molecules on the pore walls; and a hysteresis loop at intermediate relative humidities, with desorption occurring at lower pressure than adsorption. Although the nature of the hysteresis loop and its relationship to pore structure are still debated, the adsorption branch is typically interpreted as the result of capillary condensation.9,11 The close resemblance between the reflectance isotherm and mass sorption isotherms for water in porous silicon, as reported in the literature for porous silicon similar to that used here,28 suggests that there is a direct, close-to-linear relationship between adsorbed mass and reflectance, and thus that studying optical changes is a useful tool to access local concentration variations. In addition to providing spatially resolved information, the reflectance method also allows working with small pore volumes, for which it would be challenging to detect liquid uptake using traditional mass measurement techniques.29 Compared to a mass isotherm, we note additional features in reflectance isotherms. In particular, ”spikes” stand out above the hysteresis loop at the edge of desorption (for a slightly below 0.5) and in the later stages of capillary condensation (for a ≃ 0.6−0.65). We interpret the presence of these spikes as a loss of reflectance associated with increased light scattering in the porous medium. Such light scattering effects have been reported during adsorption and desorption in Vycor nanoporous glass, and have been interpreted as the sign of collective effects across a large number of pores, resulting in the formation of clusters of liquid and vapor reaching sizes comparable to the wavelength of light.30,31 Invasion Dynamics. We then recorded dynamics change in reflectance with the sample bonded to glass such as depicted in

(7)

where we have defined α′ = dα/da as the derivative of the adsorption function. eq 7 is a complex diffusion equation with activity-dependent coefficients, but for small variations in activity, it reduces to is a simple diffusion equation with an effective diffusivity

D=

k α′

(8)

eq 8 illustrates the competition between transport and adsorption: when the adsorbed mass changes rapidly as a function of vapor pressure (α′ large), adsorption on the pore walls act as a sink, slowing down the propagation of the vapor. eqs 7-8 are very general and will serve as a basis for discussion of the possible mechanisms that are at play during equilibration of the samples with vapor. The generic transport parameter k incorporates the physics of transport. If transport is dominated by diffusion in the vapor phase, by Fick’s law we have q = ϕMDm/τ(∂C/∂y), where Dm is the molecular diffusivity, M is the molar mass, and C = p/(RT) is the concentration, assuming that the vapor behaves as an ideal gas, and τ is the tortuosity of the pore space. The transport coefficient is then k = Dmpsatvm/ (τRT). In nanometer-size pores, the diffusivity can be estimated as Dm = (2rp/3) 8RT /(πM ) from Knudsen diffusion,27 yielding C

DOI: 10.1021/acs.langmuir.6b04534 Langmuir XXXX, XXX, XXX−XXX

Article

Langmuir Figure 1b, during equilibration with a vapor of constant activity a = p/psat after complete evacuation. In such a configuration, exchange of mass was only possible at the open edge (left) of the sample, and we monitored, optically, the resulting lateral propagation of changes in reflectance. At high imposed relative humidities (a > 0.6), the equilibration process resulted in a clearly visible invasion front propagating away from the open edge (Figure 3a). Image

Figure 4. Diffusion-like dynamics at low relative humidities (a < 0.6), here for a = 0.50. Raw images (inset) did not show a clear front as in Figure 3a, but image analysis (main panel) unveiled a slow, diffusionlike process. Colors represent different times of 40 min (2400 s), 80 min (4800 s), and 120 min (7200 s) after the beginning of the experiment. The dashed lines are solutions of the diffusion equation, with a constant diffusivity D = 2 × 10−10 m2/s.

imbibition speed coefficient w = Y2/t and a typical front width Δw = w+ − w− with w+ = Y+2/t and w− = Y−2/t as a function of Ψ(a). The results are shown in Figure 5 and demonstrate a

Figure 3. Imbibition-like dynamics by capillary condensation at high relative humidities (a > 0.6), here for a = 0.98. (a) Raw images obtained during the experiment (the field of view is that defined in Figure 1c). The lighter band on the left edge originated from an illumination artifact due to the proximity to the diced edge, and was excluded from the analysis. (b) Results of image analysis showing the reflectance change as a function of position. Colors represent different times, starting from t = 0 (black points around ΔI/I = 0) and with a time separation of 300 s (5 min). The horizontal dashed lines show how the front position is extracted from the data. (c) Front position squared as a function of time.

Figure 5. Front properties for imbibition-like dynamics at high imposed vapor activities (a > 0.6). (a) Imbibition speed coefficient, w, as a function of the Kelvin water potential Ψ(a) (bottom axis) calculated from the activity a (top axis) with eq 3. Values of w were extracted from experiments such as those in Figure 3, as the slope of the Y2(t) data in Figure 3c. The error bars were obtained from the values w− and w+ estimated from Y− and Y+, respectively. The lines are linear fits of w, w−, and w+. (b) Imbibition front width, Δw = w+ − w−, as a function of water potential.

analysis allowed us to extract the front profile as a function of time, after averaging over the x direction. The resulting data (Figure 3b) showed the propagation of a sharp front separating a fully wet zone (to the left, ⟨ΔI/I⟩x ≃ 0.2) from a dry zone (to the right, ⟨ΔI/I⟩x = 0). Note that the fully wet intensity was lower than that on the reflectance isotherm (0.20 vs 0.26) due to the presence of the glass layer on top of the porous silicon layer. We quantified the front propagation by extracting the mean front position Y(t) and estimated its width using Y+ and Y− (see Figure 3b). As shown in Figure 3c, the three parameters, Y, Y+, and Y−, exhibited a t1/2 scaling, typical of imbibition dynamics.9,24 For lower imposed vapor pressures (a < 0.6), the equilibration dynamics was qualitatively different, with a diffusion-like penetration of moisture into the sample rather than a welldefined front (Figure 4). Quantitatively, the dynamics was also much slower, as shown by the very different time scales in Figures 3b and 4. We performed a series of experiments in the imbibition-like regime (a > 0.6), at different imposed vapor activities, a, or equivalently different vapor water potentials Ψ(a) from eq 3. Using linear fits of the data as in Figure 3, we extracted the

linear decrease in the imbibition speed as Ψ becomes more negative, as well as a front width that is weakly dependent on the imposed vapor activity.

■

DISCUSSION In all the experiments reported above (Figures 3−5), we started with the sample in a dry state (a = 0), and the equilibration with a vapor at a > 0 resulted in moisture uptake in the pores. As a result, the processes involved in the dynamics relate to the adsorption branch of the isotherm. From the reflectance isotherm of Figure 2, it is clear that the value a = 0.6 marked a transition: for a > 0.6, capillary condensation resulted in complete filling of all the pores, while for a < 0.6 the sample was only partially saturated. Consequently, it is reasonable to assume that in dynamic situations, imposing a > 0.6 resulted in D

DOI: 10.1021/acs.langmuir.6b04534 Langmuir XXXX, XXX, XXX−XXX

Article

Langmuir

convenient way to vary the capillary stress in the pore liquid (i.e., the driving force for the flow) in a continuous manner through Kelvin equation. The parameters κ and ΔPc are associated with the viscous drag in the pores and with capillary effects, respectively, and their measurement offers an interesting way to probe the fluid mechanics of confined liquids. As we have discussed recently,10 the measured values of κ and ΔPc suggest that the viscosity and capillarity of water are not noticeably modified in nanoscale confinements: using Laplace law of capillarity (eq 4) with θ = 25 ± 5°, as measured using the sessile droplet method on porosified silicon surfaces,10,28 and the bulk surface tension of water σ = 0.073 N/m, we estimate a pore size of rp = 1.9 ± 0.2 nm; this value is compatible with pore size estimates using Kelvin-Laplace equation on the reflectance isotherm obtained here (rp = 1.8 ± 0.5 nm),36 or BJH analysis from N2 mass isotherms (rp = 1.4 ± 0.4 nm) and dynamic permation flow measurements (rp = 1.7 ± 0.2 nm) obtained on a separate sample.10 The pore size value of rp = 1.9 ± 0.2 nm is also compatible with the Carman-Kozeny relation κ = ϕreff2/(8ητ) using the bulk water viscosity η = 1.14 × 10−3 Pa·s, the tortuosity of the porous layer τ = 4.5, and an effective hydraulic radius reff = rp (1 − d/rp)2 that accounts for the immobility of one monolayer of thickness d = 0.31 nm at the pore wall.10 This consistency, as well as the compatibility of the observed imbibition dynamics with the stresses predicted from Kelvin equation, provide another demonstration of the excellent extension of the macroscopic laws of capillarity, viscous flow, and thermodynamics at the nanoscale for water.10,18 According to eq 6, when the external vapor activity a is low enough such that Ψ(a) = ΔPc, there is no net capillary force driving bulk liquid flow. From eq 2, Ψ(a) = ΔPc with the value of ΔPc determined above corresponds to a = 0.59 ± 0.02, in accordance with the experimentally observed transition to a diffusion-like regime below a ≃ 0.6. In the diffusion-like regime, while there is no capillary driving force for equilibration, there is a chemical potential driving force, as the dry sample needs to uptake some water to come to equilibrium (see Figure 2). As a result of this chemical potential imbalance, mass diffusion occurs through the porous medium, explaining the diffusionlike dynamics below a = 0.6. We estimated the effective diffusivity in that regime by fitting the data as in Figure 4 with analytical solutions of the one-dimensional diffusion equation with constant diffusivity D.37 We note that the diffusivity in eq 7 is not constant a priori, so the simple fitting with a constant D only provides a way to discuss order of magnitudes. As can be seen in Figure 4, the hypothesis of a constant diffusivity fits the data reasonably well. The diffusivity obtained from the fit was similar (D ≃ 2 × 10−10 m2/s) for two experiments in the diffusion-like regime, one below the hysteresis cycle (a = 0.35) and one in the capillary condensation regime (a = 0.50). This value may seem low compared to typical bulk molecular diffusivities, but D is in fact not a true molecular diffusivity, but rather incorporates effects from both transport coefficients in the porous medium and local adsorption thermodynamics. Indeed, molecules taken up in the pores are either adsorbed locally or transported further in the medium, as illustrated by eq 8. Based on Knudsen diffusion, eq 8 however predicts a value for D that is 2 orders of magnitude smaller than the experimental value (see Theory). This disagreement suggests that surface transport in the adsorbed water at the pore walls may play an important role in the diffusion dynamics. In fact, the observed value of D is

the saturation with liquid of the pores at the sample edge, similarly to what would happen if the sample were directly dipped in the bulk liquid, a case classically described with Lucas−Washburn dynamics. Here, however, the samples were not in contact with bulk liquid but with a subsaturated vapor; in this scenario, we expect that an additional meniscus exists at the open end of the pore, as depicted in Figure 1g. The driving force for the liquid flow thus resulted from the competition of the capillary pressure of two menisci, one at the open edge and the other at the advancing front (Figure 1g), leading to the modified Lucas-Washburn dynamics expressed in eq 6, with a natural driving force proportional to Ψ(a) − ΔPc; Ψ(a) represents the capillary pressure imposed by the local liquid− vapor Kelvin equilibrium (eq 2) at the sample edge, while ΔPc represents the intrinsic capillary pressure associated with the pore geometry and wetting properties (eq 4). The classical Washburn relation is recovered in the limit a → 1, i.e., when the vapor approaches saturation; our model thus predicts that imbibition dynamics is identical when using saturated vapor and when using bulk liquid as the boundary condition at the sample edge. Experimental results as in Figure 5a show excellent agreement with the linear w(Ψ) dependence predicted by eq 6. Furthermore, the w = 0 intercept and the slope of the linear fits in Figure 5a allows us to estimate from eq 6 both the intrinsic capillary pressure ΔPc = −70 ± 5 MPa and the Darcy permeability of the porous layer κ = 1.82 ± 0.05 m2/(Pa·s) independently, using ϕ = 0.45.10 These values are almost identical to those obtained from drying-induced permeation experiments on similar porous silicon layers that we reported recently.10 This excellent agreement validates our interpretation of the imbibition-like dynamics and the applicability of the modified Lucas−Washburn equation that takes into account the Kelvin capillary pressure (eq 6); it also suggests that our assumption that the optical signal, ΔI/I, reflects the local mass uptake in the medium was appropriate. The success of this modified Lucas−Washburn treatment (eq 6 applied to Figure 5a) indicates that imbibition from a saturated vapor (a = 1, Ψ = 0) is equivalent to imbibition from bulk liquid, when local equilibrium at the exposed edge is ensured. This equivalence has been debated in the membrane science community under the name ”Schroeder’s paradox”, based on observation of distinct absorption properties from pure liquid and saturated vapor in swelling media.32,33 Neuman and colleagues have provided an explanation of this effect as arising from history-dependent properties of certain materials (e.g., Nafion).34 For rigid membranes with well-defined pore structure, such as the porous silicon studied here, our observations support the expectation that imbibition dynamics should be defined by the chemical potential at the boundary, independent of the phase that imposes this boundary state. Classical Lucas−Washburn imbibition experiments in bulk liquids only provide a measure of the product κΔPc,18,24 except if the pressure P of the liquid reservoir in contact with the sample can be varied: in this latter case, the w(P) response allows the estimation of κ and ΔPc separately, but requires the use of large positive pressures of magnitude comparable to ΔPc.35 The technique proposed here uses a similar strategy, but the pressure at the edge is brought to large, negative values, taking advantage of the fact that small changes in vapor activity result in large changes in the Kelvin stress (eq 2). We recently reported another method exploiting Kelvin stresses to measure both κ and ΔPc, using drying-induced, steady-state permeation flows.10 Globally, using vapor−liquid equilibrium provides a E

DOI: 10.1021/acs.langmuir.6b04534 Langmuir XXXX, XXX, XXX−XXX

Article

Langmuir typical of that reported for surface flows in porous glass and porous silicon.38 Regardless of the physical mechanism behind the effective diffusion process, we note that both the imbibition-like regime and the diffusion-like regime result in invasion dynamics scaling as t1/2, resulting in equilibration times τ ∼ L2/w and τ ∼ L2/D, respectively, for a sample of dimension L. As w and D are separated by more than an order of magnitude, equilibration in the imbibition-like regime is much more efficient than in the diffusion-like regime. Also, since both processes display the same qualitative dynamics ∝ t1/2, global mass uptake experiments are not able to distinguish between them, and only spatially resolved measurements as presented here can unveil their contributions. Finally, we comment on the width of the imbibition-like front (Figures 3c and 5b): since both Y+ and Y− scaled as t1/2 (Figure 3c), the widening of the imbibition front ΔY = Y+ − Y− also followed a t1/2 scaling. The quantity Δw = w+ − w− = (Y+2 − Y−2)/t is a measure of that widening; as can be seen in Figure 5b, Δw was nearly independent of the imposed water potential and thus also insensitive to the imbibition speed in our experiments. A t1/2 scaling for front broadening during an imbibition process is considered anomalous,4 but has been reported recently in nanoporous glass;8 these authors suggested that the anomalous broadening was due to uncorrelated menisci movement during imbibition when the pores are sufficiently long compared to the spacing of lateral connections. Due to the currently limited knowledge about the local structure and connectivity of porous silicon ⟨111⟩ layers, it is not obvious whether the pores in our study satisfies these geometrical conditions or if the anomalous front broadening originates from other physics. Future work using high-resolution electron microscopy should help clarify this question.

providing useful information about the capillary and viscous effects at the nanoscale, respectively; globally, our results also demonstrate the validity of Kelvin equation in pores only ≃4 nm in diameter. For applications in, for example, sensing and membrane processes, our observations and analysis provide a robust basis for the design of materials at the pore-scale to control imbibition dynamics.

■

AUTHOR INFORMATION

Corresponding Authors

*E-mail: [email protected]. *E-mail: [email protected]. ORCID

Olivier Vincent: 0000-0002-8876-6072 Notes

The authors declare no competing financial interest.

■

ACKNOWLEDGMENTS The authors thank Glenn Swan for technical support and Antoine Robin for help in building the vacuum system. This work was supported by the National Science Foundation (IIP-1500261), the Air Force Office of Scientific Research (FA9550-15-1-0052), the U.S. Department of Agriculture (2015-67021-22844), and was performed in part at the Cornell NanoScale Facility, a member of the National Nanotechnology Infrastructure Network (National Science Foundation; Grant No. ECCS-15420819).

■

REFERENCES

(1) Stroock, A. D.; Pagay, V. V.; Zwieniecki, M. A.; Holbrook, N. M. The Physicochemical Hydrodynamics of Vascular Plants. Annu. Rev. Fluid Mech. 2014, 46, 615−642. (2) Kudrolli, A. Granular matter: sticky sand. Nat. Mater. 2008, 7, 174−175. (3) Marcolli, C. Deposition nucleation viewed as homogeneous or immersion freezing in pores and cavities. Atmos. Chem. Phys. 2014, 14, 2071−2104. (4) Alava, M.; Dubé, M.; Rost, M. Imbibition in disordered media. Adv. Phys. 2004, 53, 83−175. (5) Lucas, R. Ueber das Zeitgesetz des kapillaren Aufstiegs von Flüssigkeiten. Colloid Polym. Sci. 1918, 23, 15−22. (6) Washburn, E. W. The Dynamics of Capillary Flow. Phys. Rev. 1921, 17, 273−283. (7) Acquaroli, L. N.; Urteaga, R.; Berli, C. L. A.; Koropecki, R. R. Capillary Filling in Nanostructured Porous Silicon. Langmuir 2011, 27, 2067−2072. (8) Gruener, S.; Sadjadi, Z.; Hermes, H. E.; Kityk, A. V.; Knorr, K.; Egelhaaf, S. U.; Rieger, H.; Huber, P. Anomalous front broadening during spontaneous imbibition in a matrix with elongated pores. Proc. Natl. Acad. Sci. U. S. A. 2012, 109, 10245−10250. (9) Huber, P. Soft matter in hard confinement: phase transition thermodynamics, structure, texture, diffusion and flow in nanoporous media. J. Phys.: Condens. Matter 2015, 27, 103102. (10) Vincent, O.; Szenicer, A.; Stroock, A. D. Capillarity-driven flows at the continuum limit. Soft Matter 2016, 12, 6656−6661. (11) Charlaix, E.; Ciccotti, M. In Handbook of Nanophysics: Principles and methods; Sattler, K., Ed.; CRC Press, Boca Raton, FL, 2010; Chapter Capillary condensation in confined media, pp 12.1−12.17. (12) Bocquet, L.; Charlaix, E.; Ciliberto, S.; Crassous, J. Moistureinduced ageing in granular media and the kinetics of capillary condensation. Nature 1998, 396, 735−737. (13) Barsotti, E.; Tan, S. P.; Saraji, S.; Piri, M.; Chen, J.-H. A review on capillary condensation in nanoporous media: Implications for hydrocarbon recovery from tight reservoirs. Fuel 2016, 184, 344−361.

■

CONCLUSION We have used the optical signature induced by the uptake of water in a model nanoporous medium to study the dynamics of invasion by capillary condensation in the nanopores as a function of the imposed vapor pressure. In a broad ranges of high vapor pressures, invasion occurred as a well-defined front reminiscent of Lucas−Washburn imbibition dynamics, with an additional contribution to the capillary force driving the flow that we have shown to be well described by Kelvin equation. At lower vapor pressures, moisture transport transitioned to a slower diffusion-like behavior due to the vanishing of the net capillary driving force. Both processes (imbibition and diffusion) resulted in mass uptake proportional to t1/2, but with qualitatively different spatiotemporal dynamics, as well as orderof-magnitude difference in time scales. Our analysis also suggested that imbibition from saturated vapor was equivalent to imbibition from bulk liquid, due to the insensitivity of the dynamics to the external phase in contact with the pores. The vapor pressure dependence of the flow in the imbibition regime offers a way to finely tune the imbibition speed through the Kelvin-induced capillary pressure. This control over the dynamics by using an external, easily tunable parameter (the relative humidity of the vapor) is interesting for both fundamental studies of thermodynamics and dynamics in porous media, and for the development of technologies using the response of materials to humidity changes. For fundamental aspects, we have shown that the imbibition-like response allows an independent measurement of both the intrinsic capillary pressure and of the permeability of the porous medium, F

DOI: 10.1021/acs.langmuir.6b04534 Langmuir XXXX, XXX, XXX−XXX

Article

Langmuir (14) Wallacher, D.; Künzner, N.; Kovalev, D.; Knorr, N.; Knorr, K. Capillary Condensation in Linear Mesopores of Different Shape. Phys. Rev. Lett. 2004, 92, 195704. (15) Aubry, G. J.; Bonnet, F.; Melich, M.; Guyon, L.; Spathis, P.; Despetis, F.; Wolf, P.-E. Condensation of helium in aerogel and athermal dynamics of the random-field ising model. Phys. Rev. Lett. 2014, 113, 085301. (16) Neimark, A. V.; Ravikovitch, P. I.; Vishnyakov, A. Bridging scales from molecular simulations to classical thermodynamics: density functional theory of capillary condensation in nanopores. J. Phys.: Condens. Matter 2003, 15, 347. (17) Valiullin, R.; Naumov, S.; Galvosas, P.; Kärger, J.; Woo, H.-J.; Porcheron, F.; Monson, P. A. Exploration of molecular dynamics during transient sorption of fluids in mesoporous materials. Nature 2006, 443, 965−968. (18) Gruener, S.; Hofmann, T.; Wallacher, D.; Kityk, A. V.; Huber, P. Capillary rise of water in hydrophilic nanopores. Phys. Rev. E 2009, 79, 067301. (19) Kiepsch, S.; Pelster, R. Interplay of vapor adsorption and liquid imbibition in nanoporous Vycor glass. Phys. Rev. E: Stat. Phys., Plasmas, Fluids, Relat. Interdiscip. Top. 2016, 93, 043128. (20) Vincent, O.; Sessoms, D. A.; Huber, E. J.; Guioth, J.; Stroock, A. D. Drying by Cavitation and Poroelastic Relaxations in Porous Media with Macroscopic Pores Connected by Nanoscale Throats. Phys. Rev. Lett. 2014, 113, 134501. (21) Fisher, L.; Israelachvili, J. Determination of the capillary pressure in menisci of molecular dimensions. Chem. Phys. Lett. 1980, 76, 325− 328. (22) Gor, G. Y.; Bertinetti, L.; Bernstein, N.; Hofmann, T.; Fratzl, P.; Huber, P. Elastic response of mesoporous silicon to capillary pressures in the pores. Appl. Phys. Lett. 2015, 106, 261901. (23) The imbibition mass flux (ϕρliq /2) w/t can be compared to a typical condensation flux obtained from the kinetic theory of gases p ρliq vm/(2πRT ) , which is much greater than the imbibition flux as

(35) Li, L.; Kazoe, Y.; Mawatari, K.; Sugii, Y.; Kitamori, T. Viscosity and wetting property of water confined in extended nanospace simultaneously measured from highly-pressurized meniscus motion. J. Phys. Chem. Lett. 2012, 3, 2447−2452. (36) Kelvin−Laplace equation rp = 2σvm cos θ/[RT ln(a*)] combines Kelvin equation and Laplace equation to relate the pore size to the activity a* of the confined liquid−vapor phase transition.11 Considering that a* is located in the activity range 0.45−0.65 of the reflectance isotherm hysteresis, we estimate rp = 1.8 ± 0.5 nm. (37) Crank, J. The mathematics of diffusion; Oxford University Press: Oxford, UK, 1979. (38) Dvoyashkin, M.; Khokhlov, A.; Naumov, S.; Valiullin, R. Pulsed field gradient NMR study of surface diffusion in mesoporous adsorbents. Microporous Mesoporous Mater. 2009, 125, 58−62.

long as t ≫ 10−3 seconds, using typical values corresponding to our experiments. (24) Masoodi, R., Pillai, K. M., Eds. Wicking in Porous Materials: Traditional and Modern Modeling Approaches; CRC Press, Boca Raton, FL, 2013. (25) Abeles, B.; Chen, L. F.; Johnson, J. W.; Drake, J. Capillary condensation and surface flow in microporous Vycor glass. Isr. J. Chem. 1991, 31, 99−106. (26) Gor, G. Y. Adsorption Stress Changes the Elasticity of Liquid Argon Confined in a Nanopore. Langmuir 2014, 30, 13564−13569. (27) Gruener, S.; Huber, P. Knudsen Diffusion in Silicon Nanochannels. Phys. Rev. Lett. 2008, 100, 064502. (28) Kovacs, A.; Meister, D.; Mescheder, U. Investigation of humidity adsorption in porous silicon layers. Phys. Status Solidi A 2009, 206, 1343−1347. (29) As an example, we estimate the pore volume in our sample to be Vp ≃ 10−9 m3, corresponding to a mass change of Δm ≃ 1 mg between the dry sample and the sample saturated with water. (30) Page, J.; Liu, J.; Abeles, B.; Deckman, H.; Weitz, D. Pore-space correlations in capillary condensation in Vycor. Phys. Rev. Lett. 1993, 71, 1216. (31) Soprunyuk, V. P.; Wallacher, D.; Huber, P.; Knorr, K.; Kityk, A. V. Freezing and melting of Ar in mesopores studied by optical transmission. Phys. Rev. B: Condens. Matter Mater. Phys. 2003, 67, 144105. (32) Choi, P.; Datta, R. Sorption in Proton-Exchange Membranes An Explanation of Schroederõs Paradox. J. Electrochem. Soc. 2003, 150, E601−E607. (33) Vallieres, C.; Winkelmann, D.; Roizard, D.; Favre, E.; Scharfer, P.; Kind, M. On Schroeder’s paradox. J. Membr. Sci. 2006, 278, 357− 364. (34) Onishi, L. M.; Prausnitz, J. M.; Newman, J. Water-nafion equilibria. Absence of Schroeder’s paradox. J. Phys. Chem. B 2007, 111, 10166−10173. G

DOI: 10.1021/acs.langmuir.6b04534 Langmuir XXXX, XXX, XXX−XXX